RDFSのドメインとレンジ：グラフ文から型を導く

Once an ontology defines entities, relationships, and properties, the graph can be checked and information can be derived from it. The speaker introduces RDFS and OWL as supporting technologies placed beside the graph. They are not ordinary business entities inside the graph. They are mechanisms that read the graph's meaning and add control or inference.

A layer of rules that supports the graph

A graph contains relationships between entities. However, having a relationship does not mean that every type it implies is written explicitly. A formal layer such as RDFS or OWL reads rules about relationships. It then infers additional information from those rules.

What domain and range mean

Here, the domain is the kind of thing at the subject, or the left side of the verb. The range is the kind of thing at the object, or the right side of the verb. Therefore, domain and range are not merely labels for documentation. They can use the direction of a relationship to derive the types of entities.

The speaker explains this direction with the short graph statement “Bob teaches Scooter.” It expresses the relationship that Bob teaches Scooter. The important point here is to distinguish the left side of the statement from the right side.

Inferring types from one statement

First, the fact written directly in the graph is that Bob teaches Scooter. Next, a domain and a range are set for the relationship teaches. In the speaker's example, the domain of teaches is teacher, and its range is student.

The inference proceeds in this order.

1. Bob teaches Scooter is written directly.
2. From the domain rule that the subject of teaches is a teacher, we can infer that Bob is a teacher.
3. If there is also a rule that every teacher is a person, we can infer that Bob is a person as well.
4. From the range rule that the object of teaches is a student, we can infer that Scooter is a student.

In other words, the first statement does not have to explicitly say that Bob is a teacher, Bob is a person, and Scooter is a student. Applying the rules produces those types. The key is to keep the fact written directly in the original statement separate from the inference results added by the rules.

Do not reverse domain and range

In this example, Bob becomes a teacher because Bob is on the left, the subject of teaches. Scooter becomes a student because Scooter is on the right, the object. If domain and range are reversed, we would incorrectly infer that Bob is a student and Scooter is a teacher. When reading a relationship, first check who performs the relationship and who is its object.

Why this is more than documentation

If domain and range were only written as schema documentation, they would merely help readers understand the meaning. In the speaker's example, however, the rules actually produce new type information. That information can later be used to check whether an agent's output or a tool result fits the domain.

Connecting this to agent validation

Suppose an agent calls a tool and receives a result. The surrounding program turns that result into a form that can be validated. It then uses the ontology and rules such as those in RDFS to check whether the result fits the domain assumptions. If type inference reveals a contradiction, the program can avoid accepting the result as it is and make a decision such as sending it back to the agent for another attempt. In the next stage, OWL property characteristics and constraints also support this validation.

Bob teaches Scooter is the example shown by the speaker. The basic point about domain and range is to derive the types of the things on the left and right sides of a relationship. The goal of this chapter is to understand that direction and inference flow, without assuming detailed formal notation or a diagram.